Andrea Rotnitzky’s research centers in the development of analytical tools for estimating, from non or imperfect experimental data, the effects of interventions. This work falls into the general area of causal inference and missing and censored data analysis. She is primarily interested in the development of semiparametric efficient methods that exploit the information in the available data without making unnecessary assumptions about the parts of the data generating process that are not of substantive interest.

Most important questions in Public Health are about the effects of interventions, e.g. changing a health policy, approving new drugs or implementing optimal treatment strategies. The answers to these questions often relies on either non-experimental, i.e. observational, data or on imperfect experimental data, i.e. randomized trial data from suffering from non-compliance, drop-outs, intermittent non-response, censoring, etc.

Rotnitzky's work includes (1) Modern flexible machine learning methods for causal inference, (2) Efficient causal effect estimation in causal graphical models, (3) Estimation, from longitudinal health care databases, of the causal effects of covariate dependent treatment strategies, (4) Methods for evaluating diagnostic markers from studies that suffer from verification bias, (5) Methods for correcting for informative non-response in longitudinal studies, (6) Methods for analyzing failure time and quality of life adjusted failure time endpoints in studies with competing informative causes of censoring and, (7) Methods for analyzing clinical trials with non-compliance